About this Author
DBL%20Hendrix%20small.png College chemistry, 1983

Derek Lowe The 2002 Model

Dbl%20new%20portrait%20B%26W.png After 10 years of blogging. . .

Derek Lowe, an Arkansan by birth, got his BA from Hendrix College and his PhD in organic chemistry from Duke before spending time in Germany on a Humboldt Fellowship on his post-doc. He's worked for several major pharmaceutical companies since 1989 on drug discovery projects against schizophrenia, Alzheimer's, diabetes, osteoporosis and other diseases. To contact Derek email him directly: Twitter: Dereklowe

Chemistry and Drug Data: Drugbank
Chempedia Lab
Synthetic Pages
Organic Chemistry Portal
Not Voodoo

Chemistry and Pharma Blogs:
Org Prep Daily
The Haystack
A New Merck, Reviewed
Liberal Arts Chemistry
Electron Pusher
All Things Metathesis
C&E News Blogs
Chemiotics II
Chemical Space
Noel O'Blog
In Vivo Blog
Terra Sigilatta
BBSRC/Douglas Kell
Realizations in Biostatistics
ChemSpider Blog
Organic Chem - Education & Industry
Pharma Strategy Blog
No Name No Slogan
Practical Fragments
The Curious Wavefunction
Natural Product Man
Fragment Literature
Chemistry World Blog
Synthetic Nature
Chemistry Blog
Synthesizing Ideas
Eye on FDA
Chemical Forums
Symyx Blog
Sceptical Chymist
Lamentations on Chemistry
Computational Organic Chemistry
Mining Drugs
Henry Rzepa

Science Blogs and News:
Bad Science
The Loom
Uncertain Principles
Fierce Biotech
Blogs for Industry
Omics! Omics!
Young Female Scientist
Notional Slurry
Nobel Intent
SciTech Daily
Science Blog
Gene Expression (I)
Gene Expression (II)
Adventures in Ethics and Science
Transterrestrial Musings
Slashdot Science
Cosmic Variance
Biology News Net

Medical Blogs
DB's Medical Rants
Science-Based Medicine
Respectful Insolence
Diabetes Mine

Economics and Business
Marginal Revolution
The Volokh Conspiracy
Knowledge Problem

Politics / Current Events
Virginia Postrel
Belmont Club
Mickey Kaus

Belles Lettres
Uncouth Reflections
Arts and Letters Daily
In the Pipeline: Don't miss Derek Lowe's excellent commentary on drug discovery and the pharma industry in general at In the Pipeline

In the Pipeline

« Rating A Massive Pile of Compounds | Main | Targacept's Antidepressant Fails, And How »

November 7, 2011

Where's the Best Place to Apply Modeling to Drug Discovery?

Email This Entry

Posted by Derek

An e-mail correspondent and I were discussing this question, and I thought it would be an interesting one for everyone. He's a computational guy, and he's been wondering where the best use of computation/modeling effort in drug research might be. The obvious place to apply it is in lead generation and SAR development - but is that the best place? Is it the rate-limiting step enough of the time?

Problem is, the things that are often limiting steps are not as amenable to modeling. These are things like toxicology, target selection, and the like, and I'm not sure what they're susceptible to, except that simulation is probably not the answer. Or not yet, anyway. So what's the sweet spot, the place that maximizes importance and feasibility?

Update: an early vote for clinical trial design, which is a strong contender. Can't say that that doesn't get right to the hard part. . .

Comments (38) + TrackBacks (0) | Category: In Silico


1. drug_hunter on November 7, 2011 1:25 PM writes...

One perhaps non-obvious but very useful place is clinical trial design.

Permalink to Comment

2. PharmaHeretic on November 7, 2011 1:30 PM writes...

Oh, I don't know. It is all bullshit anyway and nobody who has money wants to discover drugs.

Permalink to Comment

3. anchor on November 7, 2011 1:46 PM writes...

Question: What is the similarity between the bogus science of astrology and computational design ?
Answer : The both offers entertainment value and have been proved wrong many times. The computational model (i.e. after the results) makes it look a better and effective presentation material for your boss to yap around in those impact meetings.

Permalink to Comment

4. johnnyboy on November 7, 2011 2:08 PM writes...

Can only speak for preclin toxicology, which is not, and I don't think will ever be, amenable to reliable modelling. We already have enough issues just trying to understand the significance of the tox findings we get. Predicting them isn't gonna happen, no matter what the computational and toxicogenomics enthusiasts would have you believe.

Permalink to Comment

5. SteveM on November 7, 2011 2:28 PM writes...

I do mathematical modeling for a living including simulation (though not on biological systems). Agree with the above generally though. A lot of modeling is GIGO obscured by visual eye candy graphics.

Moreover, for some reason modelers often lose all sight of significant digits. I.e., Systems modeled with WAG input parameters spit out results to 9 decimal places.

But the guys selling the software are in it to make a buck. I imagine on the bio-medical end too.

BTW, the worst modeling is obviously the government junk with their 20 year projections of whatever, presented deterministically with no error bands.

Permalink to Comment

6. JC on November 7, 2011 2:34 PM writes...

Making a pretty picture to explain what we already know after the fact.

Permalink to Comment

7. Ex Merck on November 7, 2011 2:43 PM writes...

The biggest impact I have seen was when our computational guys designed an excel macro to help everyone calculate their severance package.

Permalink to Comment

8. Kiru on November 7, 2011 3:10 PM writes...

Lots of experts on computational design in this comment series. Sortof reminding me of all the drug discovery experts in the software business.

Permalink to Comment

9. Curious Wavefunction on November 7, 2011 3:33 PM writes...

My comment seems to in moderation, but it's worth remembering that pretty pictures can sometimes be very useful and may reveal realistic molecular dimensions. The thing is, medicinal chemists who are used to looking at 2D depictions are usually not as familiar with the 3D sizes and shapes of molecules as are modelers. Crystallographers can also be familiar with 3D structures but they are usually too focused on the protein to look closely at the small molecule. Space-filling pictures can especially provide a very realistic view of the impact of modifications on a compound in ways that medicinal chemists who are used to 2D depictions cannot imagine (and I speak from experience). Of course as my past advisor says, the prettier the picture looks the less you should trust it, but as with everything else it has its utility.

As for software developers, it's worth noting that many of them have a very good chemical background. Of course they all have their biases (who doesn't?) and some of them do get carried away with the algorithm at the expense of the chemistry, but it would be wrong to assume that all of them know just about programming and nothing about molecular behavior. I know several drug discovery software scientists and their understanding of organic and physical chemistry is as good as anyone else's. As with any other business, it's important to know who to trust, and you cannot blame others if you swallow their opinions hook line and sinker.

Permalink to Comment

10. Boghog on November 7, 2011 3:39 PM writes...

IMHO, modeling is most useful in hit identification (virtual screening) and hit to lead where qualitative modeling can still contribute. Modeling is much less useful in lead optimization (with the possible exception of ADME and off target modeling) where quantitate predictions are needed, but are beyond the capability of current methods.

Permalink to Comment

11. UICAlchemist on November 7, 2011 3:41 PM writes...

Computational chemistry gives us pretty pictures and speculative insight into answers we already know. The most appropriate place that a good medicinal chemist would use comp chem is in evaluating basic physiochemical properties during lead optimization. Your time is better spent reading scientific literature, rather than trying to put together some fancy algorithm that will work on an ad hoc basis.

Permalink to Comment

12. Dickweed Jones on November 7, 2011 3:46 PM writes...

Best use is to get a useless paper published so you have one more useless item on your CV when the reaper comes by.

Permalink to Comment

13. CMCguy on November 7, 2011 3:47 PM writes...

If by clinical trial design mean running statistical projections, risk/event assessments and powering of the studies use of modeling is mandatory and fairly sophisticated. Are there other applications to clinical that should mention?

From process (and analytical) development there is serious benefits to Statistical Process (Method)Design to get the most out of fewer sets of experiments. Its good to always verify correlations you already believe are important but even better when points out significant interactions that did not think would have much impact.

Permalink to Comment

14. Innovorich on November 7, 2011 3:51 PM writes...

Computational chemistry, used well, can accelerate drug discovery, meaning fewer compounds are made, in less time, and have better properties at the end of the design process. This then has a direct impact on solving toxicology issues (although certainly doesn't solve them).

Computational chemists need to work closely with medicinal chemists to interactively and iteratively design compounds. Comp. chemists also need to have time, space and guidance to develop new more effective methods and roll more and more of the simpler tools out to the med. chemists.

Bioinformatics and even structure-based (where available) target drugability assessment can aid in target selection.

Permalink to Comment

15. Curious Wavefunction on November 7, 2011 4:45 PM writes...

Most modelers who give answers that are already known to other chemists (as the above commentators seem to be claiming without citing any evidence) are of the push-button type, people who treat modeling programs as black boxes without understanding their real strengths and limitations. There are even now several problems (most notably the general problem of determining solution and bioactive conformations of organic molecules) where it can be difficult or impossible to provide an answer without some kind of modeling. And no, the answer that emerges is often one that's not already known.

SteveM: Mathematical modeling per se plays a relatively minor role in drug discovery modeling. The ideal computational chemist depends as much on his intuitive knowledge of organic and physical chemistry as on knowledge of specific algorithms and statistical analysis.

Permalink to Comment

16. LeeH on November 7, 2011 4:46 PM writes...

The hardest part about using modeling in a drug discovery effort is getting the chemists to understand the importance of shifting the probabilities to your favor. It's like blackjack. Sometimes you're telling someone to hit when they have 14 and the dealer is showing 10. You might still lose, but your chance of winning goes up.

In terms of where in the process you get the most bang for the buck, it's always where the stakes are highest and the money involved is maximal. That would suggest applications late in the process, so clinical trials is sensible. But using any advantage throughout the process is also sensible.

Permalink to Comment

17. pete on November 7, 2011 4:49 PM writes...

Along the SAR line:
I'd vote that modeling can *sometimes* be useful in "proximal MOA studies". Example: competing compounds that hit a complex target in different ways.

Assuming that 1) the target is pretty well studied and 2) you can get some PK and/or PD data for the compound interactions, a model can - at a minimum - help you organize your knowledge. And, at best, it can help guide development decisions by allowing simulations that test your predictions.

Permalink to Comment

18. Pete on November 7, 2011 5:06 PM writes...

My experience is that the scope for using molecular modeling tools is greatest in lead generation. In particular we can search databases using complementary approaches such as pharmacophore-matching, molecular shape-matching and docking-n-scoring and use molecular similarity searching to ensure that different chemotypes are adequately sampled.

Lead optimisation is a different game and the modeller needs to concentrate effort where he/she can make a difference. Ideally the modeller in and LO project will also be able to function as a physical-organic chemist because the role of the modeller in LO is to help build a deeper understanding of the properties of the project molecules such as conformation, pKa and interaction potential. The modeller can also help relate properties of molecules to properties of compounds.

Permalink to Comment

19. lynn on November 7, 2011 7:11 PM writes...

Model/compute the physicochemical attributes necessary to get small molecules into the cytoplasm of Gram negative bacteria, especially Pseudomonas aeruginosa. That's certainly a rate limiting step.

Permalink to Comment

20. milkshake on November 7, 2011 8:40 PM writes...

From what I have seen, modeling in lead discovery/optimization is helpful when backed up by lots of X-crystallography work that has (in ideal case) several dissimilar ligands co-crystallized with the protein of interest. A good computational chemist will teach medicinal chemists how to look up X-ray data in PyMOL and will work with them on proposals "how to reach the back side of that left binding pocket" Comp chem is rather useless in docking studies - and virtual screening of libraries is beyond useless. You can use comp chem for generating ideas in drug design but you should not use it as a filter of ideas or structures. (Also, molecules originating from projects that too heavily depended on rational drug design tend to be overdesigned and hard to make).

Permalink to Comment

21. Anonymous on November 7, 2011 9:08 PM writes...

I've worked with modelers extensively in LO.

To Echo Milkshake, if you have lots of crystal structures and support, you can do neat things.

A niche area is the design of libraries/series. Not too much change for the molecule, and since you are throwing everything against the wall any ways, some modelling input can go a long way.

You have to ask the right questions and be willing to run with some of their ideas, even if they don't pan out. We are all doing science, so we should respect someone wiliness to want to try to validate some of their hypothesis.

My 2 cents

Permalink to Comment

22. Biostatistician on November 7, 2011 11:15 PM writes...

I'll ratify the comments made re: modeling telling an already clear story. In my experience modeling has been most useful after a finished product hits the market and thousands of patients are taking it. Before that, the drug discovery folks that I have worked with usually don't need me to tell them when something works or has bombed.

Permalink to Comment

23. Calvin on November 8, 2011 4:16 AM writes...

Well I worked with an outstanding comp chemist and my experience was that you can use it anywhere to solve problems. But largely in the sense that it's a great way to come up with new ideas that you might not otherwise see or consider. But you still need to make compounds and test the hypothesis and see where you are and refine further. It's still only part of the iterative toolbox if you like. It's rarely about clicking "calculate" and generating the solution to your problem on one go. So in that sense I found comp chemistry really useful and powerful and our comp chemist was particularly good at articulating the caveats in any piece of work. So use it but don't be dogmatic about it.

Permalink to Comment

24. vanderHolst on November 8, 2011 4:35 AM writes...

Difficult to define a specific step. Thats where I spend most of may time (based on requests).

Before MedChem gets involved:
-Searching external compound collections for in-sourcing (virtual screening 2D and 3D).
-Analysis of HTS data (clustering, filtering, IP).
-Support assay development (where to cut the protein etc).
LO, modelling is requested when the going gets rough:
-If the lead is not flat, where to introduce substituents, cycles, macrocyles etc.
-Prioritize suggestions for IP-generation (eg heteroatom permutation).
-Which property relates to permeability, tox, ADME/PK.
-How to improve specificity, avoid resistance (structure based and 2D)?
-General mathematical support, eg. Atropisomerism, PK/PD simulations, merging datasets, significance analysis...

Permalink to Comment

25. simpl on November 8, 2011 9:01 AM writes...

The best models are the ones that make unexpected predictions which turn out to have value.
I'm not surprised that toxicologists are uneasy about mathematical models. Their animal and tissues are alternative models (of humans), developed over time and shown to be useful.

How about financial models? I remember the case of naftifin, a good drug to eliminate fungi (think athlete's foot). The marketing guys came with a clear sales estimate as argument why a successor must be for oral use, which I found hard do swallow. A long search later, terbinafine was the result, and the sales predictions were right. It seems that dermatologist wanted the cure independent of analysing what the fungus was.

Permalink to Comment

26. selina on November 8, 2011 4:36 PM writes...

Diverse opinions here. Interesting.

My view:

Modern computational methods can be very useful and reliable to help elucidating structure of small molecules.

For many of the reactions with competing pathways to major/minor products, to be able to calculate the barriers for the competing steps is extremely helpful to understand the nature of the transition states and lead to new reaction design.

Permalink to Comment

27. bioguy on November 8, 2011 10:00 PM writes...

Clinical trial design and analysis, including

PK/PD modeling for dose selection
Simulated clinical trials for not only powering the study but also planning interim analyses, defining study population, endpoint selection, other design characteristics
Interpretation of results (posterior distributions, statistical significance, predictive probability, etc)
More rare biological systems-based modeling of intended PD readout

Permalink to Comment

28. Morten G on November 9, 2011 8:00 AM writes...

Biology could use some black box statistical software to determine the number of experiments to run based on estimates of noise and signal etc to achieve statistical significance. I can't remember a single biochemistry paper where they wrote that they had used statistical evaluations before they did the actual experiments. So really clinical trials methods but adapted for small labs.

Permalink to Comment

29. dmpc on November 10, 2011 11:05 AM writes...

One problem is that the modelers DON'T generally work with the chemists as tehy are sometimes kept in separate buildings. It's almost as if the chemists don't WANT the modelers at all.

The whole idea was to use this technique to save money and time yet it's not being used that way, when it easily could.

Permalink to Comment

30. chris on November 10, 2011 12:30 PM writes...

To be really useful the computational chemist has to be embedded in the chemistry team, they also need to be able to say when the computational tools are not useful for addressing a particular problem.
The list from @24 is pretty accurate.

Permalink to Comment

31. dmpc on November 10, 2011 1:10 PM writes...

One other thing I wonder: why is a modeling blog written by an organic chemist? If I dared to write about synthesis, I'd be shot!

Why can't people stick to their fields of training?

Permalink to Comment

32. ThePragmatist on November 10, 2011 5:17 PM writes...

Many interesting (and contradictory) comments. As a currently unemployed compchemist with 16 years experience I would also state that the list #24 posted is good, perhaps adding homology modeling in the absence of co-crystal structures.

Receiving difficult problems is typical. Some of these are worth tackling. Others are a waste of time. Whether the team leader or manager supports that decision is highly variable.

The PK/PD and clinical modeling is a different beast from molecular modeling. Getting biochemical and cellular "right" is tough enough. Absorption is handled with rule based methods ok, while Pgp, etc models are problematic. Good luck with predicting activity in serum as well!

The advent of automated/workflow/webserver/outsourced docking may be another source of "I told you so". It may be faster and cheaper, but better? I doubt it. As stated before: intuition and experience make a difference.

Some chemists are interested, others despise it, still others think they can do it themselves.

Permalink to Comment

33. dmpc on November 11, 2011 10:55 PM writes...

I'm also an unemployed compichemist. I have almost twelve years of experience. People don't want to pay you, but they are glad to use you for free. in the past year, I've consulted on a multiple number of projects, each time hoping a check will follow, and I wait, and nothing comes. I have enough money to pay one more month's rent, while synthetic chemists get hired for jobs I can do. Yet, I can't go the other way, even though I have five years of spectroscopy experience under my belt. I am probably one of the few people that know what VCD is.

Permalink to Comment

34. Mark Shenderovich on November 13, 2011 2:15 PM writes...

As one of the founders of molecular modeling once said, modeling does not help do get right answers,it helps to ask right questions. The best application of molecular modeling in drug design is to teach chemists and pharmacologists to think in 3D.

Permalink to Comment

35. Anonymous on November 19, 2011 9:34 AM writes...

I heard some CRO company is doing pretty decent modeling work for clients: watch out, guys, more modeling jobs will be outsourced :-)

Permalink to Comment

36. Curious Wavefunction on November 19, 2011 12:16 PM writes...

My past comment posted way earlier got lost in the approval queue because of the links so I am just going to repost it without the links.

As a modeler let me weigh in briefly. I think there are a few places where modeling can be fruitfully used. These days one of its applications is in library design. In the absence of good data it can be hard to know what the diversity of your library of small molecules is. Computational techniques can be used to measure diversity and identify "diversity holes", which can suggest new designs for the medicinal chemist. Otherwise you may end up making more of the same. The other more obvious utility is in structure-based drug design. As I described in a past post, approaches like docking can sometimes reveal counterintuitive binding orientations for leads that may save time and effort on unnecessary modifications.

However, as I detailed in a previous post, I don't think one can always identify specific time-points where modeling can be most fruitful (although one can identify time-points of the kind you mentioned where modeling is less useful). The thing is that a modeler can often bring expertise in physical organic chemistry and structural biology to the table. Such insight may sometimes be more valuable than the actual use of a technique. Modeling is more about a way of thinking (indeed, a way of life) than about any particular set of tools. It is about the ability to abstract data and experiments and structure these abstractions in a way that suggests fruitful inquiry.

Permalink to Comment

37. direwolfc on November 21, 2011 3:43 PM writes...

detecting a lot of bitterness on this comment thread...I think the point that steps that most need modeling being the least amenable to modeling is right on - and that's because those are also the toughest, most complex problems out there. Toxicology is a perfect example.

The complaints here on modeling apply to any methods applied to drug discovery - chemists synthesizing compounds that are the most tractable, not the most likely to succeed, biochemists expressing proteins that are the most well behaved, not the best therapeutic targets, etc...moral of the story is that drug discovery is hard and it's easy to trash on any part of it.

Permalink to Comment

38. direwolfc on November 21, 2011 3:44 PM writes...

detecting a lot of bitterness on this comment thread...I think the point that steps that most need modeling being the least amenable to modeling is right on - and that's because those are also the toughest, most complex problems out there. Toxicology is a perfect example.

The complaints here on modeling apply to any methods applied to drug discovery - chemists synthesizing compounds that are the most tractable, not the most likely to succeed, biochemists expressing proteins that are the most well behaved, not the best therapeutic targets, etc...moral of the story is that drug discovery is hard and it's easy to trash on any part of it.

Permalink to Comment


Remember Me?


Email this entry to:

Your email address:

Message (optional):

Gitcher SF5 Groups Right Here
Changing A Broken Science System
One and Done
The Latest Protein-Protein Compounds
Professor Fukuyama's Solvent Peaks
Novartis Gets Out of RNAi
Total Synthesis in Flow
Sweet Reason Lands On Its Face